Genetic algorithm-based system identification of active magnetic bearing system: A frequency-domain approach

Abstract

The main focus of this paper is on system identification of an active magnetic bearing system (AMB) using genetic algorithm (GA) for optimal controller design purpose. In the first step, an analytical model of the system is derived using principle of physics and taking into account both the rigid body and bending body modes of the system. In the next step, as AMB system is inherently open-loop unstable, a closed-loop system identification approach is adopted. The actual frequency response data are collected under closed-loop condition. As it is expected from the analytical model, the system has two dominant resonant frequencies which have to be accurately identified. To fit the frequency response of the system into a desired order transfer function, weight vectors are used to emphasise the resonant frequencies. Subsequently, GA is employed to search the optimal values of the required weight vectors and their corresponding scaling factors automatically in order to best fit the measured data. For verification of the proposed method, the model obtained from GA is compared with some well-known methods such as prediction error method (PEM) and subspace state space system identification (N4SID) method. Eventually, a PID controller and two notch filters are designed based on the obtained model and implemented on the actual system and the performance of the designed controller is compared with the on-board analogue controller.